Skip to content

Decide the client portal implementation and first-release boundaries #3935

Closed

Description

@devantler

🤖 Generated by the Agentic Engineer

Objective

Choose a sustainable client-portal implementation and settle the first-release boundaries before writing the application. This assessment records decisions and actionable follow-ups; it does not deploy a portal or authorize customer work.

Unified site-and-portal decision — 9 October 2026

The maintainer now requires one website and portal in devantler-tech/business-site. This supersedes the earlier separate repository/product/tenant direction. Keep the existing root Astro + Starlight public frontend; add the selected Go application boundary under server/ in that same repository. The first identity slice in #3936 builds on the working site instead of bootstrapping an empty portal repository. Share theme tokens, fonts and assets through the site's build, without creating another independently maintained frontend clone. Go remains the bounded server/identity choice; this is not a benchmarked portfolio-wide frontend winner, and #4019 still owns that measurement.

The intended runtime serves the built public pages and protected /portal routes from one canonical site entry point and one versioned deployment artifact. Keep PostgreSQL/session records, private templates, secrets and APIs outside the static output and public preview/publication artifacts. The server must authorize every protected route before rendering or returning data; a static fallback/catch-all and public caching cannot bypass that policy. Single-repository development is not permission to share client, operator or infrastructure identities.

The live Pages site is not replaced until #3939 proves the combined runtime, immutable receipt, provider configuration, private-store recovery and rollback, and #3940 verifies the complete invited customer/operator/worker path. Initially keep the portal default-off; do not add a working-looking production login link, collect real records, change DNS or enroll a paid service. These existing delivery issues remain the owners; no duplicate implementation backlog is needed. Queue and legal work live in business-site #20 and #21, preserving their native parent and Project 5 history.

Custom portal and monthly-payment decision — 9 October 2026

The maintainer has now decided that this must be our own customized portal, not an ERP, invoicing SaaS or provider-hosted customer portal substituted for the application. This settles the product ownership/UI/workflow boundary; it does not settle the frontend framework, approve custom authentication or authorize a new financial service. Reuse maintained components at narrow boundaries, especially one federated identity broker and an appropriate payment integration.

Monthly payments must be designed explicitly. Keep recurring invoices with audited operator confirmation of credited bank transfers as the initial comparison baseline; this is not automatic debit and neither banking nor hosting is assumed free. Evaluate maintained free/self-hosted billing components against any justified paid service for the actual monthly collection requirement. Record verified license terms, component versus bank/processor fees, total recurring/transaction costs, security/maintenance effort, legal/provider responsibilities, reconciliation, failure/refund/cancellation behavior, data location and exit path. Recommend a paid integration only with evidence that no suitable free/self-hosted alternative satisfies the requirements, and obtain approval of the named cost proposal before enrollment or commitment. Do not build card storage, bank/payment initiation or regulated processing infrastructure merely to avoid fees; a free SDK does not make money movement free.

Keep the portal's order authorization and monthly hosting entitlement separate. Agreement acceptance is not recurring-payment consent; a previous payment does not authorize a later period, changed scope or repeated collection. Settlement must remain independently verified server-side or through the audited operator path; browser success is never proof.

Danish/EU applicability, remediation and real-path verification are now tracked in client-portal devantler-tech/business-site#21 and business-site #19. These issues distinguish B2B/B2C, privacy, business disclosures, subscription/withdrawal/cancellation rules, PMV/tax/accounting and accessibility. Qualified legal/accounting review is required where material interpretation or facts remain uncertain. They are launch obligations for the affected features, not a claim of compliance or an excuse to bypass ordinary reviewed merges.

  • Record the settled custom-application decision, preserving federation and private customer data.
  • Record the conditional free-first monthly-payment assessment and approval-before-enrollment boundary.
  • Complete that payment comparison and record the selected maintained implementation, costs and applicable legal responsibilities before payment activation.

Settled maintainer decisions

The following choices come from the maintainer's interactive answers, including the cost-boundary clarification on 9 October 2026:

  • Both private individuals and business customers must be supported, invitation-only initially. This is the maintainer's corrected answer and supersedes the earlier business-only choice. Do not implement a company-only access or ordering restriction. Enforce invitations server-side for both audiences; implement applicable consumer agreement, withdrawal, cancellation, subscription and complaint safeguards before enabling the affected transactions. Invitation-only access does not remove those obligations; unresolved customer/transaction classification cannot be bypassed by a checkbox.
  • Application source is public in devantler-tech/business-site. The portal is a module of the unified site, not a separate application repository. Customer records, agreements, payment evidence and operator notes remain private; the existing site has no real portal implementation yet.
  • Access is invitation-only initially. Support Google and Microsoft through one federated OIDC broker, such as Dex. The portal must not implement separate provider integrations, infrastructure login or its own identity provider. Credentials and provider registration are configured privately.
  • New paid work uses an invoice and operator-confirmed bank transfer. Only an audited operator confirmation of actual credited funds for the exact order, amount, currency and accepted quote version releases work. A customer screenshot, browser return, issue label or agent assertion never does. Credited funds are not an absolute-finality or no-refunds promise.
  • Paid services may be proposed with their costs before any enrollment. There is no approved monthly paid-service budget or authorization to enroll, upgrade or collect subscription payments. Compare the existing platform baseline with any justified paid alternative, including recurring charges, usage limits, storage/backups, data location and an exit path; obtain explicit approval before a paid commitment. Existing hosting capacity is not zero-cost or verified spare capacity.
  • The website and portal form one product, repository and intended deployable site. The private server boundary and customer/operator authorization remain explicit. Keep the current static publication until the combined runtime and actual portal workflow are verified; no production cutover is authorized by this decision.
  • Hosted projects include the agreed monthly maintenance scope: optimization, accessibility and bug fixes. Features and larger iterations need a new agreement. Keep this separate from new paid-order eligibility; maintenance is not a blanket authorization for arbitrary client requests, nor a 24/7 human-support promise.

Confirmed business facts include Devantler Tech, CVR 46830385, Nikolai Emil Damm and ned@devantler.tech. Processing, retention, terms and tax wording still need appropriate review before real customer activation. Do not publish the private address or claim statutory compliance from registration alone; #3917 remains open for its remaining scope.

Bounded monthly-payment comparison — 9 October 2026

Keep the first release aligned with the already selected invoice and audited operator-confirmed bank-transfer workflow. The custom portal can manage monthly invoice periods and hosting entitlements without adopting another customer portal or committing to an additional billing SaaS. This is not automatic collection, and actual bank, hosting and operator costs are not assumed to be zero. Implementation still needs the legal, accounting, reconciliation and failure-path checks in client-portal devantler-tech/business-site#21.

  • Maintained self-hosted candidate: Kill Bill's official subscription guide describes subscription/catalog/invoice operations through an API suitable for a custom application. Its source license is Apache-2.0. It is a billing engine, not proof of free banking or payment processing; the documented server, administration and database footprint must be measured against our needs. No trial, release-specific dependency/license audit, adoption or operational-fit verdict has been completed.
  • Paid comparison only, not selected: Stripe Billing's Danish pricing lists pay-as-you-go billing at 0.7% of billing volume. Stripe Payments' Danish pricing separately lists standard EEA card processing at 1.5% + DKK 1.80; other payment, international, tax and exception costs depend on the actual setup. These are dated public list prices, not an approved cost proposal or a claim that a free library makes processing free. Preserve our own portal if a narrowly scoped integration is later justified.
  • Decision status: retain invoice/bank transfer as the first-release baseline. No evidence currently establishes that every suitable free/self-hosted alternative fails, so no paid integration or enrollment is authorized. Automatic collection, if later required, needs an exact requirements/fit comparison, verified provider eligibility and legal responsibilities, and approval of the named complete cost proposal before commitment.

This records options, not completion of the remaining stack, storage, deployment or payment-activation decisions. Keep this Spike open until its acceptance criteria and actionable follow-ups are satisfied.

Evidence and reuse/build direction

Keep a narrow application that reuses maintained identity, session and durable-storage components. Do not adopt a complete ERP or invoicing application merely to obtain login and a client dashboard, and do not invent custom authentication or payment processing.

  1. ERPNext's customer project portal provides project/task visibility. Its official container composition includes backend, frontend, websocket, worker and scheduler services. This is a broader operating footprint, and the reviewed documentation does not establish the exact accepted-version/payment/oldest-eligible-work contract. Reuse would require a demonstrated fit, not an assumption that generic tasks form the requested queue.
  2. Invoice Ninja's client portal provides contact-scoped documents, quotes, invoices and payments. That is useful invoicing functionality, but its documented quote-to-invoice flow does not establish our sequential agent-work authorization or federation requirement. No Invoice Ninja enrollment, source reuse, license conclusion or white-label purchase is authorized by this assessment.
  3. Dex brokers authentication through connectors including Google and Microsoft. A maintained provider-agnostic OIDC client can consume that one issuer; go-oidc is one Apache-2.0 Go option, not a decision to implement providers individually or proof that a Go frontend is best. Application authorization, invitations, roles and customer isolation remain the portal's responsibility.

The recommended implementation boundary is a single application service with durable restricted storage and shared business authorization for the UI and API. Keep identity federation and any later calendar or agent adapter at explicit boundaries. Exact framework/library versions, storage/deployment fit, license obligations and operating footprint must be verified at implementation selection. No benchmark has proved a portfolio-wide frontend winner: #4019 owns representative measurements, the common standard and incremental adoption. Do not rewrite the business site or choose a portal framework by popularity to close this Spike.

Bounded identity-slice implementation decision — 9 October 2026

Select a Go application with PostgreSQL for the first identity/contact slice in #3936. Keep one server-owned authorization policy for browser pages and the later API; do not introduce a second customer store or a browser-side authority model. Use server-rendered, context-escaped HTML and ordinary accessible forms for this small first slice, preserving the professional EN/DA theme. This is a scope-specific engineering decision, not a measured claim that Go templates are the best frontend across the portfolio. #4019 still owns that comparative decision and incremental adoption.

Options considered: ASP.NET Core has maintained OIDC/cookie middleware; SvelteKit supports server-side session integration and progressively enhanced forms. Both are credible alternatives, not rejected on popularity or assumed inferior performance. For the bounded first slice, the Go direction reuses the portfolio's existing Go build/release practices, needs no separate frontend service or required SPA runtime, and leaves the later API independent of the visual layer. These are architectural fit arguments; no comparative build, latency, resource or accessibility measurement has been performed.

Use maintained components rather than implementing an identity provider, token verification or session protocol:

  • One privately configured Dex-compatible issuer, consumed through go-oidc and Go's maintained OAuth2 client. Validate issuer, audience, signature, expiry, nonce and PKCE; invitations and application capabilities remain server-owned. No Google/Microsoft-specific portal SDKs.
  • SCS for opaque server-side sessions, its PostgreSQL store, and pgx for database access. Production must not use the library's default in-memory store. Session renewal, expiry, logout and revocation need actual handler/database tests.
  • PostgreSQL holds contacts, invitation consumption and callback-attempt state. Consume single-use invitations and login attempts atomically in database transactions; a session library's request-local read/delete is not proof of concurrent replay protection. Check client/operator authorization against current server state on every protected path.
  • Use the standard context-escaping HTML renderer. Browser state-changing requests require a maintained CSRF protection strategy and negative tests. Go's CrossOriginProtection documentation explicitly allows requests without Origin/fetch metadata; it is not by itself proof of complete CSRF coverage. Do not add blanket callback or API bypasses.

Licensing direction: the Go template carries Apache-2.0; go-oidc is Apache-2.0, SCS and pgx carry MIT, and PostgreSQL has its own permissive license. Verify the exact selected revisions, transitive dependencies and required notices in #3936 before publishing code. This does not relicense KSail or the packaged agent product, nor claim that a public permissive portal license prohibits competing commercial use.

Operating-footprint decision: one application service plus a dedicated logical database/least-privilege role on appropriately isolated PostgreSQL infrastructure. No Redis, message broker, billing engine, separate frontend server or new paid service is required by this identity slice. Existing capacity, database tenancy suitability, backup/restore, replica behavior and total cost remain unverified; #3939 must measure them and obtain approval of any paid-service proposal before enrollment. Pin the supported toolchain, library releases and immutable artifact in the implementation PR, and keep the slice default-off until real provider and recovery proof.

This resolves the first-slice direction, not the remaining comparative frontend, resource measurements, monthly-payment component fit or deployment/activation gates. Keep this Spike open until its remaining acceptance criteria and follow-up requirements are actually satisfied.

Work-queue ownership decision — 9 October 2026

Choose one authoritative portal-owned queue, exposed through a versioned application API; add MCP as a thin authenticated adapter only after the underlying API and business authorization are proven. This is the engineering decision for the maintainer's requested customized portal and the proposal tracked in client-portal devantler-tech/business-site#20. It is not an implemented queue, deployed MCP server or permission to execute customer work.

Reasoning: orders, quotes, client acceptance, audited credited funds, eligibility and customer-visible progress already belong to the portal. Making GitHub issues mandatory would require keeping a second work-state system consistent without adding business authority. A separate MCP-owned store would add the same duplication. GitHub therefore remains an optional allow-listed code-delivery reference; neither an issue label, a closed PR nor failed synchronization can release or complete work. Do not delete or repurpose the already prepared private client-orders repository as part of this decision.

Shared policy and API/MCP contract

  • The portal's restricted durable store owns order/version identities, original submission time, a deterministic tie-break, append-only audit/published timelines, eligibility and fenced claim leases. Client, operator and worker capabilities are separately assigned and server-checked on every interface. Never derive roles or tenant access from tool arguments or matching contact email.
  • Clients submit bounded intake, accept their own exact quote and read their own deliberately published history. Only the operator approves/requotes scope, confirms actual credited invoice funds, assigns explicit included-maintenance scope or performs audited queue overrides. Workers see only sanitized authorized briefs and safe progress references; they cannot approve, accept agreements, confirm funds or read payment/contact evidence.
  • The API's initial worker capabilities are: list eligible work; inspect a sanitized approved brief; atomically claim the oldest eligible order; renew/release the same fenced lease; and submit progress or delivery evidence for that lease. Each modifying operation needs caller-scoped idempotency, exact version/precondition checks, bounded inputs and audit attribution. Same key with changed payload, stale versions and stale fences are rejected, not treated as a new authorization.
  • Claim selection and the initial single active software-order slot are checked in one durable transaction. Re-check current scope, exact accepted agreement, audited full credited funds and lease at execution checkpoints. Restart, retries, lease loss, cancellation, changed scope and revoked/unknown authority must not duplicate delivery or authorize a stale worker. Older newly eligible work cannot automatically preempt already active work.
  • Included monthly hosting maintenance remains a separate operator-authorized entitlement for optimization, accessibility and bug fixes. It cannot be turned into another paid feature order; equally, it cannot authorize unrelated features or larger iterations. Urgent incident/security duties remain intact. Define the client-versus-internal service class and capacity policy in the reviewed owning engineer/consumer surfaces through client-portal Own the client-work queue through an API and thin MCP adapter business-site#20, rather than granting a tool permission to rewrite it.
  • For remote HTTP MCP, use maintained authorization/protocol components, validate the resource audience and keep downstream credentials separate; never forward the incoming MCP token to another API. These boundaries follow the MCP authorization specification and its security considerations. Actual broker/protocol/version interoperability still needs a tested component decision: Dex sign-in alone is not proof of worker/MCP authorization support. No provider registration or credential grant is authorized by this assessment.
  • Default-off rollout covers UI, API and MCP. No live agent connection precedes real identity, durable order-policy and recovery proof. MCP availability never bypasses a disabled portal/order gate. Customer completion requires reviewed/merged code when relevant and verified usable delivery, with deliberately sanitized evidence rather than raw agent output.

Independently reviewable delivery slices

  1. Add private client sign-in and contact onboarding #3936 delivers real invitation-only broker identity, durable contact ownership and server-side client/operator separation; no payment or worker powers.
  2. Add product requests, progress and an oldest-first operator queue #3937 delivers private durable intake, exact quote/acceptance, audited credited-bank confirmation and the operator's oldest-eligible queue, with default-off paid release and concurrency/restart tests. Reconcile its requirements with Deliver the unified Devantler Tech site and client portal #3934 before implementation.
  3. Client-portal Own the client-work queue through an API and thin MCP adapter business-site#20 first adds the narrow worker API and lease/policy negative controls; then adds the thin MCP adapter, identity/resource-boundary controls and reviewed Agentic Engineer selection integration. Keep both increments independently reviewed and default-off until their actual paths are verified. Do not create duplicate follow-up issues for these existing delivery owners.
  4. Deploy and verify the unified business site and private portal #3939 owns configured-provider deployment, durable restart/migrations, backup/restore and measured operating cost. Activate the verified portal within the unified business site #3940 enables website handoff only after real two-client/operator/worker isolation and usable delivery evidence.

Synthetic contract proof includes two clients, one operator and two competing workers; exact quote-version/payment matching, concurrent claims, restart, requote, revoked authority, stale fencing, prompt injection, cross-customer attempts and GitHub unavailability. The API-only path must pass before MCP is added. No claim that these tests, a real queue or customer delivery currently exist.

First shippable slice and delivery sequence

Priority: high maintainer-directed publication work after the business-site bridge. The audience is the operator and invited clients; the main risk reduction is server-enforced isolation and authorization before collecting real data. The existing browser prototype answers workflow/design questions only. It has no real authentication, persistent customer records, payments or deployment.

  1. Identity and private contact onboarding — Add private client sign-in and contact onboarding #3936: establish the public application source, maintenance/CI contract and reviewed artifact path, then implement invitation-only login through the single broker with durable sessions/contact ownership. Keep unavailable providers and customer intake disabled. Deliver a real, reviewable slice without building the entire order system in one PR. Test forged/expired tokens, wrong issuer/audience, callback replay, CSRF/session boundaries, revocation, email collisions and cross-customer/operator denial. Local fixtures are not real-provider or deployed proof.
  2. Durable requests and business approval — Add product requests, progress and an oldest-first operator queue #3937: free bounded intake, exact versioned quote/acceptance, audited credited-funds confirmation, original-order queue and private published timeline. Paid work stays disabled until all authorization paths pass. Use the recorded portal-owned queue decision and client-portal Own the client-work queue through an API and thin MCP adapter business-site#20's shared API/MCP contract before adding an agent executor.
  3. Customer experience — Build the bilingual Devantler Tech client-portal experience #3938: apply the professional EN/DA business theme, accessible forms and honest pending/unavailable states. Preserve Select an evidence-backed frontend standard across the portfolio #4019's measured frontend decision and avoid a speculative whole-portfolio rewrite.
  4. Unified site deployment and recovery — Deploy and verify the unified business site and private portal #3939: follow the current reviewed tenant abstraction, privately configure the broker, prove signed immutable artifacts, migrations, durable restart and backup/restore, and measure the operating footprint. Separately owned platform recovery is not this issue's authority to override.
  5. Public handoff — Activate the verified portal within the unified business site #3940: enable business-site links only after actual configured-provider login, two-client isolation and the complete authorized operator/worker path are verified at the deployed revision. A merged foundation or local mockup cannot clear this milestone.

Consultancy pricing, confirmed meeting durations/full-day definition, capacity, lead time, cancellation terms and an actual maintained iCloud adapter remain under #3980/#3981. Preserve the narrow free request/first-meeting paths; unresolved booking choices do not authorize fixed prices, a live booking button or unpaid implementation. Real paid consultancy activation waits for those decisions and verified integration.

Do not start another implementation while the portfolio's review/verification stages exceed their reviewed WIP limits. Refine the delivery slices and finish existing publication/review work; do not move started cards upstream to manufacture capacity.

Acceptance Criteria

Timebox: one focused assessment followed by specifically named decision gaps, not repeated full surveys or a speculative delivery PR. Part of #3934.

Decision completed and follow-ups retained: #3935 (comment). The unchecked payment comparison is an activation obligation owned by #3937 and business-site #21, not a delivered payment capability.

Activity

  1. added theissue type on Oct 7, 2026
  2. self-assigned this
    on Oct 9, 2026
  3. added 3 commits that reference this issue on Oct 9, 2026
  4. added a commit that references this issue on Oct 9, 2026
  5. 2 remaining items

  6. added 2 commits that reference this issue on Oct 9, 2026
  7. added a commit that references this issue on Oct 9, 2026
  8. updated Effort Medium and Priority High on Oct 9, 2026
  9. added a commit that references this issue on Oct 9, 2026
  10. devantler commented on Oct 9, 2026

    @devantler
    ContributorAuthor

    🤖 Generated by the Agentic Engineer

    First identity slice: concrete component and security decisions

    This resolves the previously unspecified CSRF choice and makes the selected session/storage direction implementable. It is a decision record for #3936, not an implemented or deployed portal. Keep the existing public Astro frontend and one Go server boundary in devantler-tech/business-site; it is not a measured portfolio-wide frontend winner (#4019).

    The version-specific baseline reviewed on 9 October 2026 is:

    Boundary Selected baseline Source license
    One broker's OIDC identity tokens go-oidc v3.21.0 Apache-2.0
    Authorization-code flow and PKCE golang.org/x/oauth2 v0.36.0 BSD-style
    PostgreSQL driver pgx v5.11.0, through its database/sql adapter MIT
    Opaque server-side sessions SCS v2.9.0 MIT
    Durable session adapter SCS's separate postgresstore module at immutable commit ab20b3feb5e9981c1f79cee8a97a289810134163 MIT
    Form/request CSRF protection nosurf v1.2.0 MIT

    Use one bounded database/sql connection pool with pgx's maintained stdlib driver for both application transactions and the SCS adapter. Do not add a second runtime driver/pool, Redis or a separate session service. The adapter is a separate Go module: resolve and lock its actual pseudo-version at the stated commit, not an invented postgresstore/v2.9.0 version. Its module declares lib/pq; an application that uses pgx must still audit the actual resolved module graph rather than claiming that declaration has disappeared. Pin a supported Go patch compatible with the portfolio's current Go 1.26 baseline, retain license notices, and review the complete resolved licenses and vulnerability results in the implementation PR. No dependency was installed or scanned in this assessment.

    Required behavior, not assumed library guarantees

    • CSRF: select nosurf's token validation together with same-origin validation. v1.2.0 is the patched release for GHSA-w9hf-35q4-vcjw. Do not use Go's cross-origin header checks alone or exempt entire portal route families. State-changing forms, login initiation and logout use protected POST requests. A narrowly defined broker callback retains transaction-bound state/nonce/PKCE and replay protection. Reject absent/forged tokens, unsafe origins and unsafe requests lacking Origin/fetch-metadata headers; retain a positive valid-form control. Configure canonical-host/proxy trust explicitly rather than trusting arbitrary forwarded headers or sibling origins.
    • Login: go-oidc's verifier explicitly leaves nonce comparison to the application. Verify issuer, signature, audience and expiry without skip switches. Bind state, nonce, PKCE verifier and redirect destination to one durable, expiring login attempt; consume it and the relevant invitation atomically. A per-request session value is not concurrent replay protection. Identity remains broker issuer plus subject, not matching email.
    • Sessions: use Secure, HttpOnly, host-only cookies with bounded expiry and explicit rotation/logout. SameSite is additional protection, not the CSRF mechanism. The PostgreSQL adapter performs session upserts and deletion; it does not supply atomic invitation consumption or an account-revocation fence. Check authoritative membership/revocation state before protected data operations and prove that a concurrent stale request cannot restore access after logout/revocation.
    • Application boundary: test these behaviors through the actual handlers and durable database, including concurrent callback/redemption, two-client/operator denial and restart persistence. Private routes/templates/data must not enter the static output or public cache. Enabled, explicit-false and entirely-unset release states must all be tested; unconfigured identity/storage remains closed.

    Operating shape remains one application artifact and a dedicated least-privilege PostgreSQL database role. Actual capacity, tenancy isolation, replicas, migrations, restore and total hosting cost remain measured gates under #3939. No paid enrollment, customer collection, provider registration, production routing or legal-compliance claim is authorized here.

    The remaining payment-fit, consultancy/calendar and deployment criteria stay with #3935, #3980/#3981, #3939 and business-site #21. This comment does not close this Spike, clear #3936's native prerequisite or move work to Ready. The complete board read also found review/verification WIP above the reviewed limits, so it does not start another implementation PR.

  11. devantler commented on Oct 10, 2026

    @devantler
    ContributorAuthor

    🤖 Generated by the Agentic Engineer

    Monthly invoicing comparison — 10 October 2026

    The first release still uses the maintainer-selected invoice and audited operator-confirmed bank-transfer path. Monthly invoicing does not, by itself, justify paid automatic collection or replacing our customized portal.

    An additional maintained candidate is Invoice Ninja behind a narrow invoicing boundary, not its customer portal as our application. Its official recurring-invoice guide supports monthly invoice generation with Auto Bill disabled. Its payment guide describes recording a received bank transfer manually. Those features are a potential workflow fit, not independent proof that funds arrived; only our audited operator path for the exact client/order, quote version, amount and currency may release work. Each hosting period remains a separate entitlement, with cancellation, refunds and failures handled explicitly.

    License caveat: the current source license is Elastic License 2.0, not Apache/MIT or an unrestricted open-source license. It limits offering substantial functionality as a hosted/managed service and preserves licensing/notices. Treat permitted internal invoicing versus customer-facing integration as an unresolved license-fit question; do not assume a paid white-label purchase cures every restriction. The official 2026 pricing update, updated May 2026, lists self-hosted white-labeling at USD 40/year and hosted Pro at USD 14/month or 140/year. These are observed list prices, not an approved cost proposal, enrollment or total operating cost. No software has been installed or paid service enrolled.

    Keep the existing Apache-2.0 Kill Bill comparison as the more extensive subscription-engine option, with its additional operating footprint still unmeasured. The first identity/contact slice does not need either billing engine. The current comparison has not shown that all suitable free/self-hosted options fail, so a paid collection service is not selected. Existing hosting, bank fees, bookkeeping, backups and operator effort are not assumed free.

    Before payment activation, #3937 and business-site#21 still need the actual invoice/period reconciliation, retries, partial/duplicate/reversed payment controls, consumer agreement/withdrawal/cancellation responsibilities and a tested maintained component choice. #3939 owns the measured capacity, operating cost and recovery evidence. Exact release/transitive license audit and integration tests remain undone; documentation fit is not implementation or Danish-law certification. #3935 and #3936's native prerequisite/WIP gates remain unchanged. No customer, provider, payment or agent-execution capability is activated by this record.

  12. added a commit that references this issue on Oct 10, 2026
  13. devantler commented on Oct 10, 2026

    @devantler
    ContributorAuthor

    🤖 Generated by the Agentic Engineer

    Identity-first decision and delivery handoff — 10 October 2026

    The first implementation remains one default-off Go module under server/ in business-site, preserving its Astro/Starlight public build and source-owned publication. Select the supported Go 1.27.1 toolchain for the implementation; retain go-oidc 3.21.0 and pgx 5.11.0 as the narrow initial dependencies. The browser/session increment retains SCS 2.9.0, the separately versioned PostgreSQL store at immutable ab20b3feb5e9981c1f79cee8a97a289810134163, and patched nosurf 1.2.0. Add components only when actually used, audit the resolved transitive graph/notices and vulnerability results before merging, and never manufacture a store version from the SCS tag.

    Fresh primary references corroborate the broker-only connector boundary, go-oidc's requirement for application nonce validation, and the PostgreSQL store's support for pgx/stdlib. They are fit evidence, not deployed-provider proof:

    The first independently reviewable foundation proves broker-token validation, durable single-use invitation/login-attempt consumption, issuer-plus-subject membership and current authorization/revocation with actual PostgreSQL concurrency/restart controls. Keep private records outside the static build. Browser login/session forms and configured Google/Microsoft proof remain required before #3936 can be declared delivered; a merged foundation is not a live portal.

    Operating footprint remains one application plus one logical PostgreSQL database, without Redis, a billing engine, a separate frontend service or paid enrollment. #3939 owns measured capacity/cost, database tenancy, backup/restore and combined-runtime publication; #4019 owns the measured portfolio frontend standard. These are explicit follow-ups, not missing implementation choices or claims that existing hosting is free.

    The remaining consultancy/calendar decisions are recorded on #3980/#3981: proposed prices and session/full-day duration are not approved; weekly capacity, lead time, buffers and cancellation terms are unconfirmed; privately authenticated iCloud adapter selection and actual calendar proof remain uncompleted. Do not publish fixed rates, promise capacity, or activate booking before those owners satisfy their acceptance criteria. The first identity foundation neither consumes a calendar nor charges for work.

    Monthly-payment comparison remains an activation obligation in #3937 and business-site #21. Invoice plus audited confirmation of credited bank funds is the selected initial boundary; no billing engine, card storage, automatic debit or paid service is adopted here. The component/fees/eligibility/accounting/legal comparison is still uncompleted and must be finished before affected payment activation.

    The maintainer explicitly approved one portal-foundation PR WIP exception in this interactive turn; ordinary ownership, security, review, CI and merge gates remain unchanged. It is not permission to start a second competing implementation, claim legal compliance, collect real records or cut over production.

    This completes the Spike's implementation decision and named delivery handoffs; transaction/calendar activation tasks remain open with their existing owners.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

Fields

Priority

High

Effort

Medium

Projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions